Large Language Models (LLMs) have revolutionized various domains, including natural language processing, data analysis, and software development, by enabling automation.In software engineering, LLM-powered coding agents have garnered significant attention due to their potential to automate complex development tasks, assist in debugging, and enhance productivity.However, existing approaches often struggle with sub-optimal decision-making, requiring either extensive manual intervention or inefficient compute scaling strategies.To improve coding agent performance, we present Dynamic Action Re-Sampling (DARS), a novel inference time compute scaling approach for coding agents, that is faster and more effective at recovering from sub-optimal decisions compared to baselines.While traditional agents either follow linear trajectories or rely on random sampling for scaling compute, our approach DARS works by branching out a trajectory at certain key decision points by taking an alternative action given the history of the trajectory and execution feedback of the previous attempt from that point.We evaluate our approach on SWE-Bench Lite benchmark, demonstrating that this scaling strategy achieves a pass@k score of 55% with Claude 3.5 Sonnet V2.Our framework achieves a pass@1 rate of 47%, outperforming state-of-the-art (SOTA) opensource frameworks. 1

Recent research on word-level confidence estimation for speech recognition systems has primarily focused on lightweight models known as Confidence Estimation Modules (CEMs), which rely on hand-engineered features derived from Automatic Speech Recognition (ASR) outputs. In contrast, we propose a novel end-to-end approach that leverages the ASR model itself (Whisper) to generate word-level confidence scores. Specifically, we introduce a method in which the Whisper model is fine-tuned to produce scalar confidence scores given an audio input and its corresponding hypothesis transcript. Our experiments demonstrate that the fine-tuned Whisper-tiny model, comparable in size to a strong CEM baseline, achieves similar performance on the in-domain dataset and surpasses the CEM baseline on eight out-of-domain datasets, whereas the fine-tuned Whisper-large model consistently outperforms the CEM baseline by a substantial margin across all the datasets.

In this work, we explore regions as a potential visual analogue of words for self-supervised image representation learning. Inspired by Masked Autoencoding (MAE), a generative pre-training baseline, we propose masked region autoencoding to learn from groups of pixels or regions. Specifically, we design an architecture which efficiently addresses the one-to-many mapping between images and regions, while being highly effective especially with high-quality regions. When integrated with MAE, our approach (R-MAE) demonstrates consistent improvements across various pre-training datasets and downstream detection and segmentation benchmarks, with negligible computational overheads. Beyond the quantitative evaluation, our analysis indicates the models pre-trained with masked region autoencoding unlock the potential for interactive segmentation. The code is provided at https://github.com/facebookresearch/r-mae.

Whisper, despite being trained on 680 K hours of web-scaled audio data, faces difficulty in recognizing rare words like domain-specific terms, with a solution being contextual biasing through prompting. To improve upon this method, in this paper, we propose a supervised learning strategy to fine-tune Whisper for contextual biasing instruction. We demonstrate that by using only 670 hours of Common Voice English set for fine-tuning, our model generalizes to 11 diverse opensource English datasets, achieving a 45.6% improvement in recognition of rare words and 60.8% improvement in recognition of words unseen during fine-tuning over the baseline method. Surprisingly, our model’s contextual biasing ability generalizes even to languages unseen during fine-tuning.

This paper revisits the standard pretrain-then-finetune paradigm used in computer vision for visual recognition tasks. Typically, state-of-the-art foundation models are pretrained using large scale (weakly) supervised datasets with billions of images. We introduce an additional pre-pretraining stage that is simple and uses the self-supervised MAE technique to initialize the model. While MAE has only been shown to scale with the size of models, we find that it scales with the size of the training dataset as well. Thus, our MAE-based pre-pretraining scales with both model and data size making it applicable for training foundation models. Pre-pretraining consistently improves both the model convergence and the downstream transfer performance across a range of model scales (millions to billions of parameters), and dataset sizes (millions to billions of images). We measure the effectiveness of pre-pretraining on 10 different visual recognition tasks spanning image classification, video recognition, object detection, low-shot classification and zero-shot recognition. Our largest model achieves new state-of-the-art results on iNaturalist-18 (91.3%), 1-shot ImageNet-1k (62.1%), and zero-shot transfer on Food-101 (96.2%). Our study reveals that model initialization plays a significant role, even for web-scale pretraining with billions of images.

Modern hierarchical vision transformers have added several vision-specific components in the pursuit of supervised classification performance. While these components lead to effective accuracies and attractive FLOP counts, the added complexity actually makes these transformers slower than their vanilla ViT counterparts. In this paper, we argue that this additional bulk is unnecessary. By pretraining with a strong visual pretext task (MAE), we can strip out all the bells-and-whistles from a state-of-the-art multi-stage vision transformer without losing accuracy. In the process, we create Hiera, an extremely simple hierarchical vision transformer that is more accurate than previous models while being significantly faster both at inference and during training. We evaluate Hiera on a variety of tasks for image and video recognition. Our code and models are available at https://github.com/facebookresearch/hiera.

Sorting is a key computational kernel in many big data applications. Most sorting implementations focus on a specific input size, record width, and hardware configuration. This has created a wide array of sorters that are optimized only to a narrow application domain.In this work we show that merge trees can be implemented on FPGAs to offer state-of-the-art performance over many problem sizes. We introduce a novel merge tree architecture and develop Bonsai, an adaptive sorting solution that takes into consideration the off-chip memory bandwidth and the amount of on-chip resources to optimize sorting time. FPGA programmability allows us to leverage Bonsai to quickly implement the optimal merge tree configuration for any problem size and memory hierarchy.Using Bonsai, we develop a state-of-the-art sorter which specifically targets DRAM-scale sorting on AWS EC2 F1 instances. For 4-32 GB array size, our implementation has a minimum of 2.3x, 1.3x, 1.2x and up to 2.5x, 3.7x, 1.3x speedup over the best designs on CPUs, FPGAs, and GPUs, respectively. Our design exhibits 3.3x better bandwidth-efficiency compared to the best previous sorting implementations. Finally, we demonstrate that Bonsai can tune our design over a wide range of problem sizes(megabyte to terabyte) and memory hierarchies including DDR DRAMs, high-bandwidth memories (HBMs) and solid-state disks (SSDs).